Sistema de contagem, classificação de grupo de idade e segmentação de vestimenta de pessoas em vídeos utilizando deep learning

Abstract

Due to the growth in the number of commercial establishments and artificial intelligence and vision methods, there has been interest in the application of such methods to understand the behavior of customers and the market, aiming to sell more and/or sell better. The process of automating obtaining information, such as age group and clothing, can be very relevant from an economic point of view. Through the analysis of this data, a commercial establishment can obtain interesting information about its target audience in such a way as to boost sales and generate monetary return by directing the establishment’s content to this audience. This work proposes an end-to-end system to help obtain relevant information from the target audience. Computer vision and deep learning methods are used to perform people counting, age group classification, as well as segmentation and clothing classification of people in videos. For this, the StrongSORT algorithm was used together with YOLO-v5 for the tracking of people, as well as methods of detection and facial recognition, in addition to face alignment to remove noise from the background. The modules were implemented on a complex distributed communication and processing system using Apache Kafka. For the persistence of the data, a MongoDB database was used, which provides the information consumed in the last stage of the system, which is the visualization of the processed data in a dashboard in a friendly way for the end user. At the end of the work, very satisfactory results were obtained in all proposed modules with different boundary conditions applied to them, in addition to resulting in a system with a robust and easily scalable data architecture.